On 11/07/2016 02:17 AM, Benjamin Gaignard wrote:
Hello Laura,
What are plumbers outputs for ION after your talk ?
Regards, Benjamin
We agreed to suspend work on new major Ion features and work towards a more generic design being pushed by the graphics team (see https://github.com/cubanismo/allocator). Platform/ devicetree support is more interesting because it will still be needed for a more generic design but the existing bindings still aren't acceptable to the DT maintainers. I don't see much value in trying to convert another board to a set of bindings that aren't actually acceptable or even refactor existing code.
The objections to the bindings are that they are still too linux specific and not platform specific enough. Most of the reason why I went with this proposal had to do with getting a device for CMA, otherwise there isn't much of a need for anything in devicetree. I'll be giving this more thought, if you have your own ideas, please feel free to submit them for review.
Thanks, Laura
2016-10-27 1:25 GMT+02:00 Laura Abbott labbott@redhat.com:
On 10/26/2016 08:05 AM, Benjamin Gaignard wrote:
2016-10-26 16:44 GMT+02:00 Sumit Semwal sumit.semwal@linaro.org:
On 26 October 2016 at 20:11, Benjamin Gaignard benjamin.gaignard@linaro.org wrote:
2016-10-26 15:51 GMT+02:00 Sumit Semwal sumit.semwal@linaro.org:
Hello Benjamin,
On 26 October 2016 at 19:02, Benjamin Gaignard benjamin.gaignard@linaro.org wrote: > > It is more or less a copy of Hisilicon driver but with a heap > definition > fitting with STIH4xx SoC needs. > I have just chnage the some function prefix from "hi6220" to "sti". > Thanks for your patches!
I was just wondering if you couldn't convert the HiSilicon driver into something like a 'simple-ion' driver, and have just the DT definitions as specifics? This would save a lot of code duplication, and keep it as a simple interface for common heaps like cma.
Create a simple-ion driver is a good idea but it means that heaps (configuration, name, etc..) will have to be describe into devicetree. I'm not sure if that will is acceptable.
If there are any ST-specific requirements that are incompatible with the existing driver, it should be clearly documented out here I think.
heaps names and Ids aren't the same so I can't reuse hisilicon driver.
But I'd suspect both these are solvable with using something like 'generic,cma' instead of 'hisi,cma' or 'st,cma'?
yes, but it requires to describe the heaps in devicetree. Hisilicon driver was doing like that until last month but Laura convert it to common platform:
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/drive...
I bet she got good reasons do to that so I have implemented sti ION driver in this mindset.
I agree that having common code would be useful. There are some generic bindings listed in drivers/staging/android/ion/devicetree.txt so we could go with linux,ion-heap-dma.
The changes got merged and there were never mailing list objections but that's because the devicetree maintainers got busy and never actually looked at them and Arnd at least still didn't like the idea of Ion in devicetree. I'd like to wait until after plumbers next week to decide what to do.
Best, Sumit.
Thanks, Laura